The $2B AI Startup Simulating Human Behavior with 85% Accuracy

Elon Musk thinks there could be a probability that we live in a simulation.

You know, I— I do think we live in a simulation. So Simile is a company that is creating a foundation model of human behavior, partnership with over 40 million real people. For the first time, really, in the field, we could actually show that we can predict people's behavior with 85%+ accuracy.

You have raised 300 million in total at a valuation of 2 billion reported. You have really great names, like I mentioned: Greenoaks, Index. So what happens if someone comes to you and offers you $15 billion to start a company?

Well—

Joon, welcome to Sourcery.

Thank you for having me.

I know we're here in Index's office. Someone who is a big contributor or investor in your company, congratulations on the recent large announcements. Well, today I'm, like, super excited. This is an area I don't really quite understand. I don't really understand where the company is positioned in the market, like what you're replacing, what you're now making.

Like, I don't know exactly what simulations are, but I find it fascinating. Elon Musk thinks there could be a probability that we live in a simulation. So what does that mean to you?

Well, you know, I— I do think we live in a simulation.

You do?

I do.

Explain.

Well, and of course I— not surprisingly, because of the company that we're building and the technology that we're building, we do get asked this question quite frequently. And at some point I started to just say, "Hey, no, we live in a simulation."

But I think the part that actually does matter more to me is, at the end of the day, is the experience that we're having actually real. Does it feel real to us? Is this true to who we are?

I think at the end of the day, that's what matters. Probability-wise, I think it's a pretty good argument that humans and societies like ours will always have a desire to better understand itself. And what we found is the best way to understand something is to recreate it.

And one way I sometimes describe what simulation is, is it's a little bit like creating a portrait of who we are as a society. So I think this is actually a very innate desire for humanity or civilization like ours to have.

In which case, then yes, any civilization would create a simulation, and at some point it probably, you know, it likely is the case that we live in a simulation.

So how in real world are we in a simulation? Like, what would make that up? And then we'll get into what you're building.

Right, and so let me say a little more about what would— what makes up the simulation. What do you mean?

Like, how could humans be living in a simulation? Just like we all have the same— we're all programmed the same way, we all have the same, like, psychological behaviors, like it's just— everything is kind of already written. Like, what does it mean?

So the way I often think about a society or an experience like what we are havingright now is there's always this quantum building block of people's tastes and behaviors. So it might actually be things like my inherent characteristics, people's risk tolerance, not really known to change.

There are certain aesthetics that I really enjoy. And these all come together, and we build individual building blocks of these larger societies, which is individuals. So really, when we think about simulations, we think about individuals first. Can we create simulations of individuals that faithfully and accurately represent a diverse persona and characteristics and behaviors of that person?

And then to create a civilization-wise simulation, that basically means we put all these people into a same environment, let them interact with each other. And the part that's always complex and chaotic is what actually gets born in their interactions.

So that's ultimately, I think, what, you know, building simulation here in the societal scale really is going to look like.

And so this all started with Smallville?

It did.

Okay. Explain what that is and, like, your whole background behind that.

For sure. So Smallville was an experiment that we ran back in 2023 and 2022, where we had the initial observation that these large language models— initially we were using GPT-3 and then GPT-3.5. The models were still fairly janky, but you could see a glimpse of opportunity where these models were trained on such broad human data that when you poke at theright angle, you can actually extract a lot of human data that appeared in its training data.

This might be from the web, Wikipedia, some of the social media. And what we really started to ask ourselves is, can we actually use this to create an interaction paradigm that is truly new, that goes beyond just simply generating context or classifying different tasks, but can we actually generate truly new behaviors?

And we decided that the most ambitious version of this is actually to create a simulation of an entire person. And then once we had an entire person simulated, can we actually go create an entire community? So Smallville is an experiment that came out of that effort.

So it was a game town. Computer scientists love games because it's an amazing way to demonstrate new capabilities. So we created a game town that looks a lot like what Pokémon, Sims, like these kind of towns might look like.

And then we populated this game town with 25 agents. And these agents all had different names, different characteristics, different life stories, and they were basically language models paired with what we called back then an agentic architecture that had memory, planning, and reflection.

So they would actually wake up in the morning, do their routines, go to work, have relationships. The simulation itself actually sets in place, takes place the day before Valentine's Day. So these agents will actually come together to have party.

So there is a character that's actually organizing Valentine's Day party. There are characters that are asking themselves out. And none of this was pre-programmed, but what we could see is that we can basically create simulations of these people with emergent phenomena just coming out of the simulation.

So that was the experiment that we ran in 2023 that was Smallville. That particular work ended up introducing a few key concepts. One was that we can create simulations of people. And the other was the core agentic architecture and that we can create agents using generative AI models.

And it was very early days of these agentic workflows. So we could really start to see them becoming possible. And then we've been building on that technology, basically for us to then create simulations that are more and more accurate, and that's more diverse, and that's bigger in scale.

How do you test the validity of this? Like, how do you know if it's actually working?

So our ability to test actually has shifted quite a bit also alongside the technological progress. On day one, when we created Smallville, what we actually did was we interviewed them. We interviewed the agents. What we've done is we pulled individual agents out of this game town, sat them in this hypothetical chair on a command line, and we would basically ask them questions.

And these questions would probe certain qualities of these agents that's similar to how people would reason about different tasks and their lives. So simple question: "Tell me your name." And then we would ask, "Tell us your life story."

We would then ask questions like, "Who's the one person you want to spend time with?" And then we would then ask, "If you had to get that person a present, what would you get them?" We would ask these questions because these are questions that when you think about human brain and human mind, we can only answer these questions because we can remember, because we can plan ahead, but also because we can reflect.

Then what we ended up doing was we generated all these responses, and then we actually had human evaluators come in and take a look at these responses and said, "Given the experience that the agents had in their game town experience, are these experiences or insights actually human-like?"

So we would actually basically have human evaluators compare between human authors' generation and these agents' responses to see which one's more realistic. And we've actually done quite well. These agents were actually quite realistic. And the way to sometimes tell is, imagine you're asking this agent, "Who do you want to hang out with the most?"

It's a little bit strange if the agent says whoever they happen to see the most on the street. The agent should likely say the person they actually care about. Who do you have a special relationship with? What kind of conversations did you have with people?

What actually stood out? Then that agent should really reflect on that pick and pick those people. So those are some of the experiments that we ran, and those are the signals that we were able to get. But then over time, we wanted to go beyond that.

So here, really, the initial evaluation really was about believability, that we started to see what we call the illusion of life in these agents. But as we matured as a field, what we started to do was to run evaluation that is now about more of a faithfulness and accuracy.

So the experiment that we ended up running following that was we ran an experiment where we then took this architecture to create simulations of 1,000 real people. So that started with us actually going out, recruiting a large number of real people that sampled representatively from the US.

So it's 1,000 representative sample of the US population across gender, race, ethnicity, and so forth. And then we actually collected data from these people. So we would actually do, this time, human interview. So we actually built an AI agent that can talk to voice-to-voice with people, have those agents talk to people voice-to-voice.

We would ask questions that's a little bit like the questions we ask agents in Smallville. "Tell me the story of your life. Where did you grow up? What were some of the hard decisions you had to make? Who matters to you?"

We would use that data to then train AI agents that can basically be the twins of these people. And then we would send people away for two weeks, bring them back to a lab, and then this time we would actually ask people to complete a battery of surveys, behavioral studies, respond to stimuli, and had their agents predict what they would say.

And at the end of 2024, first, for the first time really in the field, we could actually show that we can predict people's behavior with 85% plus accuracy with these agents. That's ultimately what gave the company foundation for today's technology.

Of course, the technology itself has shifted quite a bit, but that was the initial building block that we could see that, ah, these models and these agents can now actually predict human behavior, and this is an amazing building block.

But at the same time, it's a large data operation that we do not simply create simulations out of vacuum. We believe strongly that the simulation has to be grounded, and it has to be evaluated alongside people. So it really is both.

I have so many questions. Do you sometimes inject, like, anomalies or things that could create some factors that are not predictable? And then my second question, which I wanted to be my first question, is, is human behavior predictable?

In the training data, we actually do inject a lot of behavioral knowledge. It's not just about what they say, but it's also really about what they do. And there's a lot of irrational behaviors in human tendencies. So a lot of things that we study in economic behavior game, randomized control trials, they all try to cover those.

At the same time, there are certain things in people that's really not known to change. Your risk tolerance, as I briefly mentioned, it's not really known to change in a person. So once we are able to learn that, then this actually becomes a trait that becomes very predictable.

So it really is about understanding the gap between what people think they are and what they actually do in their life. And really the modeling challenge here actually is really learning that translational layer. How do you take observed behaviors or attitudes into their fundamental tendencies?

And what is in that gap? Is it time? Is it events around you? Is it, like, what—like, how do you determine that gap that ultimately creates the outcome?

So the part that is actually very consistent in our experimentation is these two functions are actually governed by different function or different sort of, you know, shape of the curve. What people say in survey actually often just follows a normal distribution,right?

So when I say, you know, "Am I likely to go to the gym this week?" I likely say with some normal distribution curve, you know, yes, maybe, you know, a seven-point scale, I likely say five to six every week.

Now, do I actually go to the gym that week? The activation energy for me to actually go to the gym is actually way higher than what I would expect. So that looks much more like an exponential curve where it has a much clearer drop-off in terms of what kind of activation we need to actually get to that state of, "Oh, I'm actually going to exercise this week."

So those are the kind of different functions that we need to learn, and that's where often the gap is.

Again, I have so many questions about this. What is the number one or the most common indicator of human behavior? Like, what are those traits or things that you've noticed create the reactions or events? What causes people to do things?

Why would someone make a decision versus not? Like, what are those—how do you determine if someone will continue a pattern or not?

So it turns out what this actually has to do sometimes, actually, is the existence of optionality. So let's actually briefly take a look at how psychologists would usually study human behavior in a survey. We have an abstract question, and we have basically a certain set of options that were presented in multiple-choice questions or on a likely scale of one to seven.

That's a very confined space. But when people behave, they don't usually behave in vacuum or in abstract conditions. They often behave in the real world,right? And what that basically means is even simple questions like, "Where do we want to go eat?"

In abstract, if somebody said, "Hey, let's go eat X," the odds are if I'm really hungry, in fact, if I'm very hungry and if there's no food available, I'd almost certainly say, "Yes, let's go do that." But what often ends up happening is we do not have access to certain options, but we have access to other options.

This is where people have this optionality. So they now, they have to think. They have to think about the opportunity cost of taking that option. So that's actually what really governs people's behavior, that, "Yes, I can go exercise this week, but if I don't, I can sleep for one more hour or I can take one more meeting."

It's these kind of things. Optionality and the type of optionality often really governs people's behaviors.

Do you pull a lot of data from dating apps?

Not dating apps in particular, no.

No. Okay. So taking all of this into consideration, how does this then become the business of Simile?

So Simile is a company that is creating a foundation model of human behavior that basically can take as input a description of a person or a group of people and then simulate what these people might do in different contexts.

So today, simple use cases would actually go into things like surveys, experiments. Many of the insights team have a lot of studies they were meant to run. Now what Simile can do is actually run them, but in a much faster pace, with much less resources.

So that's obvious use cases where we do concept testing. And in more complex cases, we do conjoint analysis, mixed-depth analysis. The traditional analysis that researchers had to run today that would have taken them sometimes hundreds of thousands of dollars, sometimes millions of dollars over many years, we can run them in simulation basically almost instantly.

They are just a quarry away. So that's the use cases today. But at the same time, the users then come to us and ask, "Okay, it's great that we can now do these studies. Can we then do something that our studies couldn't do?

What would people actually do? How would they actually react to different stimuli?" So this is where simulation really is about also creating a model of the environment, not just the people. So imagine that we have customers. Just dumb example here.

Imagine you're trying to sell water. It's actually quite interesting. If this as a behavior question is actually quite interesting, imagine there's only one source of water, then that particular person can likely charge us a million dollar to just have access to water.

But the reason why that cannot happen is because of the optionality, because there is a context around the water source where it is everywhere,right? It's about packaging. It's about, of course, theright stream. But at the end of the day, this is widely available, that there are other optionalities in this space.

So we want to actually—we want to model that. We want to model that environment. What are the other options? What are the other alternatives? Given that as the context, can we then expose our agents to this decision-making? At what price would you actually buy this water, given all these environmental factors that go into this?

So it really is about modeling people and their behavior, but it's also about modeling the environment and everything that surrounds them.

This episode is brought to you by Brex, my favorite. You become what you spend on. And I refuse to spend my time on work that shouldn't exist. Expense reports, receipt chasing, and manual closes. The companies building what's next from Vercel, OpenAI, Anthropic, Granola, and Deepgram all made the same call.

They all run on Brex. Brex is the intelligent finance platform that combines cards, expenses, and banking into a single stack with agentic finance built in. AI agents that handle expenses automatically, enforce policy before spend happens, and close your books in minutes.

That's why Sourcery runs on Brex, so I can spend time on building and not busywork. It's time to get Brex AF. Learn more at brex.com/sourcery. That's B-R-E-X.com/S-O-U-R-C-E-R-Y. Bye. From handshake to signature faster. Harvey, the AI platform built for legal work, is trusted by 80% of the Am Law 100 and over 20% of the Fortune 500.

Learn more today at harvey.ai. That's harvey.ai. AI needs more than chips. It needs power, land, and infrastructure. Zone develops next-generation data center campuses, partnering with AI companies, site developers, and technology leaders to bring compute online faster and at scale.

Zone is building the foundation of the AI frontier. Visit zonefrontier.com to learn more. That's zonefrontier.com to learn more. And who are some of your customers? What types of companies are these?

Yeah. So we work across a few different verticals today. Right now, one of the larger verticals is retailers.

Okay.

But one of the customers that we've been working with is, of course, CVS. And they have worked with us to create simulations of both how to better understand their consumers so that they can better serve them, both in terms of the store layout, the messaging, and the overall decor.

But we also work with financial companies. So think some of the largest financial institutions today, but we also work with international institutions like Banco Itaú in Brazil, where they also have similar needs, but they want to understand, well, what kind of policies or what kind of messaging do we need to put out to better communicate our values to our customers?

And of course, we also work with entertainment companies, CPG companies. So the verticals are pretty broad, but what unites them is their need that they need to better understand their user base, that they need to be, and they're trying to be even more customer-centric.

And today, there's a real limitation to be able to talk to every single people every time they need to talk to them. But if we can create simulations of people and be able to ask any questions at any time, that becomes a superpower for these companies.

For each one of these customers, do you create their own digital twins? Like, how do you create net new populations within that instead of using the same base for everything where you'd get duplicates?

Right. So Simile is maintaining vertical-specific agent bank across the key verticals. So these are actually simulated panels that's always available on day one. So when we get into a new customer relationship, that's always available. Now, then our customers often have further asks,right?

Because no one's actually interested in general population. Everyone's interested in their own unique population that they're trying to better understand. So in those contexts, Simile would actually go collect more data to create and train the model to be able to cover those populations.

And today, Simile have partnership with over 40 million real people across the panel partnerships that we have. This is across the globe. And at any given moment, we're talking to tens of thousands, if not hundreds of thousands, of real people every single week.

So that's a scale of operation that we are running today. So if there's a new need for new panel members, that can usually be met within 24 hours.

Who are these people? I'm always really curious about this because I don't understand polling. I don't understand when people say it's 53% of Americans think X or Y. It's like, who are these Americans? Who? I don't know. I've never met someone who's been polled before or surveyed.

Have you? Well, you have, so.

But yeah. So like, who are these people?

Yeah. So the way this operates, so these are people, like just everyday people, who answers to, who chooses, obviously, to answer the polling or decide to participate in these data collection processes. The magic to this process often is, however, is knowing how to sample them.

So social scientists have been developing methods for randomized samples, probability sampling, to really get a broad understanding of the representative set of people. In some cases, some of these panel companies, they specialize in this exercise. And what that means for them is they actually try to form a real relationship with people across the entire country.

One of the strategic partnerships that Simile has actually is with Gallup. Gallup, of course, one of the most, I would consider Gallup to be one of the most gold standard polling company or company that is generating these kind of insights.

They have a Gallup panel of hundreds of thousands of people with whom they've actually built up relationship over the past decade, two decades, and so forth.

But wouldn't those people be biased at some point or feel entitled to, like, give them what they think would be, like, the best information based on their own circumstances or something? Like, how do you actually know that that's valid information?

Yeah. So we control for these. A couple of ways we control. One, obviously, is, and the biases actually can creep up in different ways. So there are different ways to fight for different kinds of biases. And, you know, not surprisingly, this entire thing is an area of study.

Like, it's an entire academic field with many researchers working on this. But for instance, one of the clear biases that we often see is fatigue. If you get asked the same questions or even different questions, but if you get polled every single week, they start to really get fatigued, and they start to answer in ways or they try to game the system,right?

Oh, I want to, if I participate in this, I get paid more. So I just give some bullshit answer, and I just participate. You start to see those behaviors. So what you do is you actually time it so that the participation is few in between and sparse enough while being able to still maintain that relationship.

And we would actually run studies that basically shows, we have a new population versus population who had answered X number of questions over the past. What is theright amount of distance to have between asking different questions to get the high quality, what we would consider to be the gold standard, gold standard quality responses?

We would actually measure that. Another thing that we would do is, oftentimes in these kind of studies, we also do attention checks. So these are basically questions that are dummy questions that are not actually a part of the core measurement.

Yeah.

But you're basically sneaking in questions that would filter out participants who are not actually paying attention.

What would one of those questions be?

Question B, a question might be something like, imagineright after our interview, you asked me, what was the first conversation, what was the first question that you asked me in today's conversation? If I can't answer, that's tricky. You asked me about Elon.

Are we living in a simulation question? I think I should know that if I was actually paying attention to our conversation. But that's an example of attention check.

I was thinking, like, something's really silly. Checking for someone's attention. Like, how tall is a giraffe? How many hamburgers can a human eat?

So technically, that's actually also good attention checks.

Really?

Yeah.

You can use those if you want.

Thank you. This truly, I mean, but these are actually questions that we would sometimes ask too. Some surveys or data collection process would literally ask people to do very simple arithmetics so that, like, you know, if they say 10 plus 10 is 30, then we know they're likely not paying attention.

Or they're not.

Or

they might not be good at math in those instances.

That brings up another question. I don't know how to, I'm just, I'm now in the mood. But are there certain populations you have to avoid? Like drug addicts, mentally ill people, like people that might skew the data in one way or another?

Or is that part of the overall graph? I know that's like an inappropriate question, but I'm curious, like, what are the no-gos in who you survey and how do you profile that out?

So for us, the no-gos are actually participants who cannot give informed consent. So we take informed consent very seriously, that every member of panel who joined our study and our data collection process, they know why the data is getting collected, at least broadly, and they consent to that process.

And they also have the opportunity to withdraw their consent. So that's something that we take seriously. So for instance, underaged

population cannot participate. So that's actually one blind spot that we own, that we do not collect data from underaged population. And this would also generalize to other populations that cannot give informed consent. Now, that said, aside from that population, where Simile is different is we are trying to create a model that can truly be a representative snapshot of who we are as a society.

So there's actually no population that we should avoid. We actually want to know every single population that's out there. And this is also where, for instance, we're also quite different from simply creating a model, like LanguageModel or other modeling companies.

Other modeling companies, what they are trying to create is an amazing reasoning machine that's great at coding, mathematics, and fundamentally trying to be a good chatbot, good assistant. For us, we are trying to create models that can actually model the diverse viewpoints, the values, preferences, and tastes of people.

A part of it is also then learning the good behaviors, but also bad behaviors of people. Simile needs to be able to capture those.

For part of the prep work for this, we had a Simile task, what do you call it? Project? Pilot? For Sourcery. And I thought this was really interesting because I'm like, who are these people?

Yeah.

Who are you interviewing? But there were two different tests. And so one was on the topic appeal and one was on titles. And so the number one title that it guessed, or I guess, I don't know, what are the terms that you use that it, like, predicted?

Simulated.

Simulate, of course.

The number one title that it simulated was the $2 billion company building AI simulations of real people. That's the strongest title overall with my audience, which is mostly investors and, like, tech executives and that kind of thing, which was interesting.

And then the topics really went into digital twins. The number one one was digital twins, simulated customer testing, and privacy. There was no clear winner among them. I kind of agree with, becauseright now the biggest concern with all this AI stuff is cybersecurity and privacy risks.

I obviously asked a lot about, like, how simulated customer testing works. The digital twins thing is quite interesting. I don't know. I don't know. How does this work for other people? So, like, what is the process at which this, like, outputs this information?

What our users have today is they have access to an online, it's a, we basically serve a SaaS platform to our users. So they get access to this web interface where they can basically come in and define two things.

They can say who they want to talk to and what questions they want to ask. And really, they can ask anything. Andright now, they can ask any number of questions. There is no limit to the number of questions they can ask.

And we'll generate responses. That's what they get today. Now, one of the things that is the most important though is this is a new way of making decisions. We haven't quite seen simulation that is so generalizable in social context that can inform all human decisions.

So one of the things that we do, and we work very closely with our users, is the change management, helping users take these results and then helping them calibrate when to use it, how to use it, where do we need more data, how do we validate it.

One of the core functions attached to these reports in relation to what I'm just saying is this idea of model of confidence or predicted accuracy. So this actually, model is actually quite important to us. So Simile will be generating predictions, simulations of future.

Alongside the core foundation model that we are training to make the simulation though, we are jointly training a predictive model of accuracy

where its job is one thing. Its job is to output 0 to 100 range probability on how confident we are in this particular response on the basis of all the data that we use to train our model and the questions that we got.

And this is sort of like in inferential statistics, if you're a social scientist, it's a little bit like having p-value. And in science, when p-value is less than 0.05, that's sort of your canonical way of saying, ah, we're ready to make a decision.

We provide that. So if confidence is higher than 90% or usually 85%, our customers have found it to be robust evidence that can actually be used to make very high-stakes decisions. So it is actually the whole package. So it is about delivering the insight and then also delivering evidence that is undeniable that, well, if this is the evidence, then this is a totally reasonable outcome that we have to prepare for and then providing this model that basically is giving us a degree of confidence calibration.

It's really interesting. One of the questions I had was, what is the most expensive mistake Simile helped a customer avoid?

Yeah.

What is it?

So there's actually an interesting study that we actually ran. This technically was slightly, this was before Simile, but I still mention it because I think as an example, I think this is quite interesting.

We worked with a few automobile companies. And one of them, and in fact, all of them were interested in electric vehicles. Should we do it? When should we do it? How should we do it? That was the top of their mind for every single electric vehicles company that we've met.

And these were some of the largest players in the field. And one of the simulations that we ran actually had to do with, can we predict or can we simulate how people would react to different messaging for the entry of new electric vehicle in this particular market?

And how can we optimize that messaging so we convey our values to our customers? So there was a strategy that said, OK, this is a clear winner. Fantastic. We're ready to go. And this is good. What was interesting though is this actually had downstream implication.

This particular messaging, this particular approach towards the market actually shifted people's overall perception of non-electric vehicle and globally actually lowered the sales within that entire market.

Geez.

So, and obviously this was run in studies and this ended up becoming an input, but this was, again, this wasn't a part of Simile back then. But it's these kind of findings that I actually find to be quite interesting.

These are the findings that are not just narrowly focused on one particular question, but it's more generalizable. But you're looking at the entire field holistically. Simulation allow you to explore that. And when it's at its best, it actually should surface insights that's counterintuitive, that's surprising.

So that's really the goal and that's really the value.

How good is AI at real professional work? Mercor's AI productivity index, or APEX, tracks it with every major frontier model release. Build with partners like Harvey and Cognition. See where the frontier actually stands at Mercor.com/Apex. Today's episode is sponsored by VCX by Fundrise, the public ticker for private tech, allowing investors of all sizes to invest in venture capital.

Learn more at getvcx.com. Public is a multi-asset investing platform where you can automate your investing with AI agents. Create your own custom agent with a simple prompt. Agents can monitor markets, execute trading strategies, and manage your cash across stocks, bonds, options, crypto, and more.

Want to buy at market close and sell the next day's open? Just tell your agent. Automate your investing at public.com. Founders scale faster on Deel. Set up payroll for any country in minutes. Hire anyone anywhere. Get visas handled fast and get back to building.

Visit deel.com/sourcery. That's D-E-E-L.com/sourcery. Do you have any, like, dream scenarios or dream simulations that you want to answer or uncover? Like, what's the wildest, you probably have some wild things in your head. Like, what is the wildest kind of thing that you're trying to, you wish someone would want you to solve or you could solve on your own?

We're running this company and we're delivering so much value, but there's a small part of me that still is that researcher part. And as a researcher, I am inspired by questions that are truly societal and generational. Every generation has a few of those questions.

One of my research heroes is Thomas Schelling, who was a behavioral economist back in the '70s and '80s, who introduced what we called the model of segregation. And racial segregation was a big topic. That was the generational topic.

So he decided to apply simulation to solve that, to at least have an understanding of its cause and mechanisms. So I do think about, well, what are those questions that we have today? Well, one of them is climate change.

Climate change is hard, not just because of the physics of it, but because it's about coordination. There are so many actors with competing incentives who have their own things they're trying to optimize for. When they come together and clash, the equilibrium state they generate is nearly impossible to navigate.

So if we can simulate that over and over again and see what kind of outcome we might get depending on what policy, that could be potentially quite powerful. Another one is, well, if we believe AGI is here or if it is imminent, then obviously we, as a community, at least in some pockets of this community, have presented a few different ideas.

One of them might be UBI. There might be some other ideas. Can we simulate these kind of findings? Instead of running a five-year study on UBI, can we actually have simulated that and shown that this is where it worked, but this is where it didn't?

Those are good simulations to run.

Have you simulated who you should be taking money from? I know you just raised a huge round.

Well.

You must have,right?

In fact, what I often do is I do simulate these kind of conversations. So, like, obviously, you know, Molly, you have a very big online presence. So, of course, you bet I simulated the conversations before walking in.

Is this one of them?

Well, OK, here's the.

Tell me.

The funny part. It did warn me that you might ask whether we live in a simulation.

Of course.

So I should have taken that advice more seriously.

Pull up your results. Let's go through thisright now. What else did it simulate?

So that was a big one. And I thought that was, it's not something, and this is interesting,right? Because that's not something that every interviewer asks. But for some reason, this is actually a good question that the simulator actually thought, given everything that we know about you that you might ask.

Well, I do tweet at Elon a lot. So that is part of it. I am curious though, you have raised 300 million in total at a valuation of 2 billion reported. And you have really great names, like I mentioned, Greenoaks, Index.

You also have some really cool individuals like Fei-Fei Li and Andrej Karpathy. How did you collect such an interesting base of investors and angels?

Right. So we were in a way fortunate to have worked with many of these people in the past. So many of them were actually our colleagues. So Fei-Fei, of course, a close colleague at Stanford. One of my co-founders, Michael Bernstein, worked closely with her for many years.

And even today, we are colleagues. Andrej, one of my other co-founders, Percy Leung, was actually on his doctoral dissertation committee. So we've known Andrej for a long time. And in fact, when I first published the Generated Agents paper back in 2023, Andrej actually was one of the first people to really amplify that particular work.

So in that way, this sort of felt like we're coming full circle. But I think in many ways, we were also fortunate because the idea, the core research, and the core need for this technology is so real that I truly do believe that simulation in many ways is inevitable, that there's this deep innate desire and need for people to better understand who they are.

And just like how we write autographs, autobiographs, or we paint portraits, simulation is a lot like that. We are, in many ways, painting a portrait of who we are as a society. And I think many of our angels see that, that this is an inspiring idea.

This is an idea that does need to exist. And that's, I think, what got them excited.

Who was it? It was Alexander Wang was talking about how when he was first exposed to OpenClaw, I think from Nat Friedman, the prompt that Nat gave him was to, like, do a deep dive psychological analysis based on all of the information.

And OpenClaw just, like, ingested everything off of their computers, photos, like, emails, everything like that, and gave, like, a detailed psychological analysis. And Alexander Wang was shocked by it. Is that an angle as well? Like, how do you see how much of the data you need?

So it is both how much, but it's about more about what data you need. So it turns out asking theright questions actually matters quite a bit. So we do interviews with people, just like what we're doingright now. In fact, where I actually got the idea of doing interviews with people to learn about them was back in 2023 and 2024.

I actually was on a lot of podcasts for the academic work that we led at Stanford. And I realized that after talking to these interviewers for about an hour or two, they know a lot about you, especially the good ones.

If they ask theright questions, they know a tremendous amount about both your field and who you are. So that's where actually we got the inspiration from. So this is where we ask questions that are both topical, but also quite general.

We found that asking people to just tell us the story of their life actually matters a lot. It reveals a lot of interesting pockets of knowledge about those people. But we also collect a lot of behavior data. Behavior data might be observational in part.

It's like transaction data, mobility data. Makes a lot of sense. But what really matters is actually the cause and mechanism. It's the randomized control trials. Imagine you have a same experimental setup and you're varying a particular variable, whether it's a product feature, whether it's a messaging, and actually observing people's behavior change.

That data actually matters a tremendous amount because what we've learned over time is people are actually not that interested in predicting the future, which is, to me, that was a little bit surprising. But unless you're trying to win against the stock market, what are you going to do if I told you that your market is going to crash in two quarters?

They're all going to then ask, well, that's shitty. What do we need to do now to change that? But it's always the second part that matters more. So what do we need to do now to shape the future?

And this is where, for us, it's not just about finding correlation. It's about understanding the cause or mechanism, the causation, because that's ultimately the lever that we have to shift the direction of the future.

So who makes up the team? What kind of talent are you looking for? Is it researchers? Is it scientists? Is it engineers? Who's, like, how is it composed?

Right. So we are now a team of about 100 people. 20% of the team actually is our lab mates and our academic relatives. So these were people who were my lab mates in Percy's group, in Michael's group, in the computer science department at Stanford, who had then gone on to, you know, frontier labs, then coming back to work with us to build these foundation models of human behavior.

But we also have an amazing group of people in that category and within sort of our relative circles who are much more of methodologists as well. It's not just about creating models for us. It's actually about studying people, understanding people.

So we actually do have a small but growing set of methodologists who have backgrounds more from social sciences as well. So that's the science part of simulation. So about 20% to 25% of the company, but modeling, eval, and the science part of conducting these simulations, that's a very important part of this company.

And then we have an amazing group of engineers and people in the product side because this technology is getting handed to many of the Fortune 100s and other users today. So we are creating this technology with this amazing group of people.

We actually recently invited our head of eng, Patan, who came initially actually from DoorDash. And there, he was an amazing leader. He recently.

I heard, by the way, I heard DoorDash has amazing AI talent, like, very underrated.

They're killers.

Yeah.

The thing I always appreciate about DoorDash and that particular team is they're soft-spoken in some ways, but they fought in an extremely brutally competitive market and they won. That speaks volume. So we have an amazing group of engineers and product talent.

And this is now matched with an amazing group of go-to-market and people on the deployment side. We are very forward deployed as a company. So that part is led by one of my co-founders, Laney Yellen. But our fundamental thesis of this company is one.

The data, the model is extremely important. Simulation is basically in its GPT-4, GPT-4.5 era, where the technology is here to make impact, but this technology is changing and evolving every single month. And a lot of this battle is also going to be in the market in our forward deployed motion.

And this is something that I look to the previous sort of cycles in the recent AI cycle. For instance, I look at coding, agent decoding era. I actually have found that companies that were forward deployed actually found a lot of success.

Their relationship became stickier, became much more defensible over time. And I think that's also going to repeat in simulation because I really do see the simulation space as one that's going to reach its maturity to a point where, in about a year and a half to two years, we'll be where coding agent as an industry is today.

So I think there's a lot of learning that we're also taking from that era.

We're also in an era of spies and people leaking information and flight risk employees. What just happened with they had a spy or someone bugged them? Yeah. So how do you help HR teams try to determine who to hire, who not to hire?

So these are some of the use cases that we get and we do, in some cases, collaborate. And of course, we do this with informed consent and autonomy, again, from the members of our customers' companies. But some of the use cases that we hear is many of the companies, especially as they get large, they're deeply interested in retention.

How can we make sure that our members are excited and especially our most performant members are committed to the company? And that is something that we can help simulate. Another one is they also want to recruit. So imagine that now you are a company that is trying to recruit the most talented students from Stanford, MIT.

What is theright way to communicate to that population? That is also something that we can help simulate. So it's not just about understanding the consumers or the customers. It's really about understanding everyone that's in our ecosystem. And many of those important stakeholders are actually internal.

I have to say, the more we talk and the more you answer questions, the more nervous I'm getting. And I feel like you can read my mind a little bit. I don't know if that is a common occurrence that you have, but I genuinely am like, shit, you did your research.

What does the simulation say? OK, so as we close out, this is one of my favorite questions. You kind of already answered it, but I'm worried, like, I'm curious, maybe you can maybe you have a different answer. But one of my favorite questions is our partner question with Brex, which might have showed up in your simulation, but it's all about performance.

So Brex is about performance, spending smarter, moving faster. I take this in a personal bend. So I think that's a matter of who you surround yourself with. Some people say it's your five closest friends. You might have a more scientific answer to that, but it could be a mentor.

It could be someone who's inspired you along the way. I know you were also a painter, so maybe it's an artist. I don't know. But I am curious, who is that person for you or those people?

So for me, the people I respect the most and I love the most in my career actually were people with two traits. One is, are they the common denominator of success? Are these people who, in different eras of their career, stood up to basically say, I will make this successful no matter how everything else goes.

I will be the reason why this will be successful? That always, for me, demonstrated their ability to reinvent themselves and take deep ownership. And the other one is people who have superpowers, but two superpowers that's really not supposed to coexist in the same person.

These are people who I sometimes describe one of my co-founders, Michael, this way, where he is a very mature scientist, understands all about the scientific methodology and so forth, but at the same time, he's a child who gets excited by creating game demo.

Or Laney Yellen, who is deeply paranoid on day-to-day basis that we have to do everythingright now and really make sure that we do everything we can to make this successful, but at the same time, deeply religious that believes that no matter what happens, we will win and we will succeed.

Those kind of tension I found to be particularly beautiful. And of course, Percy, famously within Stanford community, my last co-founder, he's a pianist, but he's also one of the most well-known AI researchers. Amazing combination. So these people really do inspire me.

And, you know, for people who are the common denominator of success, you can see that. You can feel in their behavior, in their career. I was actually recently coming back from a workshop, and I happened to sit next to a scholar named Sandy Pentland.

He used to be a long he used to be at MIT Media Lab as a professor for a long time, actually trained many of the most brilliant minds in my field and in the computer sciences more broadly. He also has had this kind of amazing career.

And this is somebody that I always look up to, where he had many generations where there was an era where he was a physicist. There was an era when he was an AI researcher and ended up training some of the most well-known AI researchers of that generation.

There was an era when he was a computational social scientist. And in each of the eras, you could see that Sandy was such an influential character. That's also an amazing power to observe.

And what makes that up? Like, how did those people become that different person in each era? It's just extreme commitment and durability. Like, what is it?

I think it's extreme commitment, stamina, the ability to continuously get excited. And these are also people who are looking who's in the game so that they can play the next game. These are people whose greatest prize in winning the current game is gaining theright and invitation to play the next game.

That's what makes them special. And also, that's something that we try to emulate.

If those are the traits of success, what are the traits of failure? What causes people to fail?

You know, it's a I think, as they say, every family that doesn't work out have their own reasons. I think every failure can look slightly differently. But I think the reverse of what I said, I think, can sometimes be modes of failures, that you're trying to overly optimize for that one particular outcome.

One thing that I often saw was there is that one particular fundraiser you want to run. There is that one customer that you're trying to run against. There is that one best paper that you want to win. And you win it, and you go, now what?

And they lose interest. And this is where I've seen so many institutions, individuals who have achieved incredible things at some point in their career and then lose interest and over time disappear. And whether that is failure or not, that's not for me to say.

In many of those cases, I think it could actually be seen as an amazing success. But for those who want to play this game for a long time and have impact across many generations and many eras, that is a failure mode.

So it's the difference between short-term thinking and long-term thinking. Fascinating.

And why are we doing this?

Yeah.

Are we doing this so we can have an exit tomorrow? Or are we doing it because we want to fundamentally reshape how our society operates for the better?

So what happens if someone comes to you and offers you $15 billion to sell your company?

Well, if I have everyright, every reasons to believe that there is an ambitious idea that only simulate can pursue, then we're going to pursue that.

OK. Well, Joon, this was such a fun conversation. Thank you very much. And I appreciate you answering my silly questions.

Thank you for having me.

Hey, it's Molly. If you enjoy our interviews, check out our newsletter, Sourcery.vc, where we deliver a once-a-week top deals and tech headlines email and also go deeper on our podcast interviews. Subscribe to Sourcery today. And don't forget to subscribe to the podcast on YouTube, Spotify, Apple, or wherever you listen.

Link in description to sign up.